when is lying the right choice?...conversational sensing with cnl [braines et al., どでとに,...
TRANSCRIPT
when is lying the right choice?xiv • iv • mmxv.
Federico Cerutti† • Artemis Parvizi† • Alice Toniolo† • Dave Braines‡ • Geeth R. de Mel* •Timothy J. Norman† • Nir Oren† • Jeff Z. Pan† • Gavin Pearson△ • Stephen D. Pipes‡ • Paul Sullivan♢
† U. Aberdeen • ‡ Emerging Technologies, IBM • *T. J. 4atson Research Center, IBM • △DSTL • ♢INTELPOINT, Inc.
ControlledQuery Evaluation
UncertainKB
NLInterfaces
ControlledQuery Evaluation
UncertainKB
NLInterfaces
4ater Contaminated⟨で.ば, で.で, で.と⟩
..
Definition: Binomial/SL Opinion [Jøsang, どででと]
.
A binomial opinion — or SL opinion — about a proposition φ iswφ = ⟨b(φ),d(φ),u(φ)⟩, where b(φ) is the belief about φ — thesummation of the probability masses that entail φ; d(φ) is thedisbelief about φ; u(φ) is the uncertainty about φ; and b(φ) +d(φ) + u(φ) = と.
.
4ater Contaminated⟨で.は, で.で, で.ど⟩
⟨で.は, で.と, で.と⟩
.
.
4ater Contaminated⟨で.ね, で.と, で.な⟩
⟨で.ね, で.ど, で.ど⟩
⟨で.は, で.ど, で.で⟩
.
.
.
4ater Contaminated⟨で.ね, で.と, で.な⟩
⟨で.ね, で.ど, で.ど⟩
⟨で.は, で.ど, で.で⟩ ⟨と.で, で.で, で.で⟩
p-prov-o [Idika et al., どでとな]
ItalianWaterSensor
CollectWater
Samples
CollectedWater
Samples
WaterContamination
Report
WaterContamination
WasProbDerivedFrom WasProbDerivedFrom
ReportGeneration
LabWaterTesting
UKChemical
Lab
NGOChemical
Lab
Data ofNormality
Water
WasProbGeneratedByWasProbGeneratedBy
ProbUsed
ProbUsed
WasProbAssociatedWith WasProbAssociatedWith
WasProbGeneratedBy
WasProbAssociatedWith
ProbUsed
s-dl-lite kb [Şensoy et al., どでとな]
..
Definition
.
A DL-lite knowledge base K = ⟨T ,A⟩ consists of a TBox T and an ABox A. Axiomsare either
∙ class inclusion axioms: B ⊑ C ∈ T where B is a basic class B := A | ∃R | ∃R− (Adenotes a named class, R a named property, and R− the inverse of R) and C isa general class C := B | ¬B | Cと ⊓ Cど; or
∙ individual axioms: B(a),R(a,b) ∈ A where a and b are named individuals.SDL-Lite is an extension of DL-lite with subjective opinion assertions of the formB : w where w is an opinion and B is an ABox axiom.
s-dl-lite kb [Şensoy et al., どでとな]
Syntax Semantics⊤ ⊤I(o) = ⟨と, で, で⟩⊥ ⊥I(o) = ⟨で, と, で⟩∃R b((∃R)I(oと)) ≥ max
∪∀oど{b(R
I(oと,oど))} andd((∃R)I(oと)) ≤ min
∪∀oど{d(R
I(oと,oど))}¬B (¬B)I(o) = ¬BI(o)R− (R−)
I(oど,oと) = RI(oと,oど)
Bと ⊑ Bど ∀o ∈ ∆I ,b(BとI(o)) ≤ b(Bど
I(o)) andd(Bど
I(o)) ≤ d(BとI(o))
Bと ⊑ ¬Bど ∀o ∈ ∆I ,b(BとI(o)) ≤ d(Bど
I(o)) andb(Bど
I(o)) ≤ d(BとI(o))
B(a) : w b(w) ≤ b(BI(aI)) and d(w) ≤ d(BI(aI))R(a,b) : w b(w) ≤ b(RI(aI ,bI)) and d(w) ≤ d(RI(aI ,bI))
proposal: s-dl-lite provenance
ItalianWaterSensor
CollectWater
Samples
CollectedWater
Samples
WaterContamination
Report
WaterContamination
WasProbDerivedFrom WasProbDerivedFrom
ReportGeneration
LabWaterTesting
UKChemical
Lab
NGOChemical
Lab
Data ofNormality
Water
WasProbGeneratedByWasProbGeneratedBy
ProbUsed
ProbUsed
WasProbAssociatedWith WasProbAssociatedWith
WasProbGeneratedBy
WasProbAssociatedWith
ProbUsed
⟨0.6, 0.2, 0.2⟩ ⟨0.8, 0.2, 0.0⟩
proposal: s-dl-lite provenance
. . .
E(Collected 4ater Samples) : ⟨と.で, で.で, で.で⟩
E(4ater Contamination Report) : ⟨と.で, で.で, で.で⟩
RDer(4ater Contamination Report, Collected 4ater Samples) : ⟨で.ね, で.ど, で.ど⟩
. . .
ControlledQuery Evaluation
UncertainKB
NLInterfaces
4ater Contaminated : ⟨で.ね, で.と, で.な⟩
…
RU Collected 4ater SamplesNGO Lab 4ater Testingで は で と で と
ItalianWaterSensor
CollectWater
Samples
CollectedWater
Samples
WaterContamination
Report
WaterContamination
WasProbDerivedFrom WasProbDerivedFrom
ReportGeneration
LabWaterTesting
UKChemical
Lab
NGOChemical
Lab
Data ofNormality
Water
WasProbGeneratedByWasProbGeneratedBy
ProbUsed
ProbUsed
WasProbAssociatedWith WasProbAssociatedWith
WasProbGeneratedBy
WasProbAssociatedWith
ProbUsed⟨0.8, 0.1, 0.1⟩
⟨0.6, 0.2, 0.2⟩ ⟨0.8, 0.2, 0.0⟩
why 4ater Contaminated で ね で と で な ?
…
Y RU Collected 4ater Samples Y w YAg NGO Lab 4ater Testing b w で ぬ?
4ater Contaminated : ⟨で.ね, で.と, で.な⟩
…
RU Collected 4ater SamplesNGO Lab 4ater Testingで は で と で と
ItalianWaterSensor
CollectWater
Samples
CollectedWater
Samples
WaterContamination
Report
WaterContamination
WasProbDerivedFrom WasProbDerivedFrom
ReportGeneration
LabWaterTesting
UKChemical
Lab
NGOChemical
Lab
Data ofNormality
Water
WasProbGeneratedByWasProbGeneratedBy
ProbUsed
ProbUsed
WasProbAssociatedWith WasProbAssociatedWith
WasProbGeneratedBy
WasProbAssociatedWith
ProbUsed⟨0.8, 0.1, 0.1⟩
⟨0.6, 0.2, 0.2⟩ ⟨0.8, 0.2, 0.0⟩
why 4ater Contaminated : ⟨で.ね, で.と, で.な⟩?
…
Y RU Collected 4ater Samples Y w YAg NGO Lab 4ater Testing b w で ぬ?
4ater Contaminated : ⟨で.ね, で.と, で.な⟩
…
RU Collected 4ater SamplesNGO Lab 4ater Testingで は で と で と
ItalianWaterSensor
CollectWater
Samples
CollectedWater
Samples
WaterContamination
Report
WaterContamination
WasProbDerivedFrom WasProbDerivedFrom
ReportGeneration
LabWaterTesting
UKChemical
Lab
NGOChemical
Lab
Data ofNormality
Water
WasProbGeneratedByWasProbGeneratedBy
ProbUsed
ProbUsed
WasProbAssociatedWith WasProbAssociatedWith
WasProbGeneratedBy
WasProbAssociatedWith
ProbUsed⟨0.8, 0.1, 0.1⟩
⟨0.6, 0.2, 0.2⟩ ⟨0.8, 0.2, 0.0⟩
why 4ater Contaminated : ⟨で.ね, で.と, で.な⟩?
…
Y RU Collected 4ater Samples Y w YAg NGO Lab 4ater Testing b w で ぬ?
4ater Contaminated : ⟨で.ね, で.と, で.な⟩
…
RU(Collected 4ater Samples,NGO Lab 4ater Testing) :⟨で.は, で.と, で.と⟩
ItalianWaterSensor
CollectWater
Samples
CollectedWater
Samples
WaterContamination
Report
WaterContamination
WasProbDerivedFrom WasProbDerivedFrom
ReportGeneration
LabWaterTesting
UKChemical
Lab
NGOChemical
Lab
Data ofNormality
Water
WasProbGeneratedByWasProbGeneratedBy
ProbUsed
ProbUsed
WasProbAssociatedWith WasProbAssociatedWith
WasProbGeneratedBy
WasProbAssociatedWith
ProbUsed⟨0.8, 0.1, 0.1⟩
⟨0.6, 0.2, 0.2⟩ ⟨0.8, 0.2, 0.0⟩
why 4ater Contaminated : ⟨で.ね, で.と, で.な⟩?
…
Y RU Collected 4ater Samples Y w YAg NGO Lab 4ater Testing b w で ぬ?
4ater Contaminated : ⟨で.ね, で.と, で.な⟩
…
RU(Collected 4ater Samples,NGO Lab 4ater Testing) :⟨で.は, で.と, で.と⟩
ItalianWaterSensor
CollectWater
Samples
CollectedWater
Samples
WaterContamination
Report
WaterContamination
WasProbDerivedFrom WasProbDerivedFrom
ReportGeneration
LabWaterTesting
UKChemical
Lab
NGOChemical
Lab
Data ofNormality
Water
WasProbGeneratedByWasProbGeneratedBy
ProbUsed
ProbUsed
WasProbAssociatedWith WasProbAssociatedWith
WasProbGeneratedBy
WasProbAssociatedWith
ProbUsed⟨0.8, 0.1, 0.1⟩
⟨0.6, 0.2, 0.2⟩ ⟨0.8, 0.2, 0.0⟩
why 4ater Contaminated : ⟨で.ね, で.と, で.な⟩?
…
∃Y,RU(Collected 4ater Samples, Y) : w ∧ Y ̸=
Ag(NGO Lab 4ater Testing) ∧ b(w) > で.ぬ?
∃Y,RU(Collected 4ater Samples, Y) : w ∧ Y ̸= Ag(NGO Lab 4ater Testing) ∧ b(w) > で.ぬ?
4hat if:
R·(·,NGO Lab 4ater Testing) : ⟨で.で, で.は, で.ど⟩
controlled query evaluation [Biskup et al., どでとに]
(Postulated) 4orld 3iew: ⟨Γ;φと, . . . , φn⟩
Γ ∈ どLΓ background/collateral knowledge, biases, …
φi ∈ L information exchanged
Belief operator: Bel : どLΓ × L∗ 7→ どL
LΓ extends the propositional language L with rules
ψ ∈ Bel(⟨Γ;φと, . . . , φn⟩) ▶ is assumed that ψ can be believed
Secrecy policy: ⟨ψ,Bel⟩
Desire to avoid that an agent believes ψ using the operator Bel
proposal: subjective logic based ceq
と. From propositional language to S-DL-Lite KBs
ど. Strategies for CEQ (including white lies)
ItalianWaterSensor
CollectWater
Samples
CollectedWater
Samples
WaterContamination
Report
WaterContamination
WasProbDerivedFrom WasProbDerivedFrom
ReportGeneration
LabWaterTesting
UKChemical
Lab
NGOChemical
Lab
Data ofNormality
Water
WasProbGeneratedByWasProbGeneratedBy
ProbUsed
ProbUsed
WasProbAssociatedWith WasProbAssociatedWith
WasProbGeneratedBy
WasProbAssociatedWith
ProbUsed⟨0.8, 0.1, 0.1⟩
⟨0.6, 0.2, 0.2⟩ ⟨0.8, 0.2, 0.0⟩
▶
ItalianWaterSensor
CollectWater
Samples
CollectedWater
Samples
WaterContamination
Report
WaterContamination
WasProbDerivedFrom WasProbDerivedFrom
ReportGeneration
LabWaterTesting
UKChemical
Lab
NGOChemical
Lab
Data ofNormality
Water
WasProbGeneratedByWasProbGeneratedBy
ProbUsed
ProbUsed
WasProbAssociatedWith WasProbAssociatedWith
WasProbGeneratedBy
WasProbAssociatedWith
ProbUsed⟨0.8, 0.1, 0.1⟩
⟨0.6, 0.2, 0.2⟩ ⟨0.8, 0.2, 0.0⟩
ProbUsed⟨0.6, 0.3, 0.1⟩
ControlledQuery Evaluation
UncertainKB
NLInterfaces
conversational sensing with cnl [Braines et al., どでとに, Preece et al., どでとに]
Data sources Analytic services Decision maker
Courtesy of Alun Preece and Dave Braines
conversational sensing with cnl [Braines et al., どでとに, Preece et al., どでとに]
ask/tell
why
gist/expand
NL to CNL
CNL to CNL
CNL to NL
confirm
Courtesy of Alun Preece and Dave Braines
conversational sensing with cnl [Braines et al., どでとに, Preece et al., どでとに]
Hi, UK Patrol. How can Ihelp?
Moira:
UK Patrol:
Suspicious vehicle headingsouth: black saloon with
license plate ABC123
21:36
21:37
You said? vehicle, movingthing: registration is
ABC123, direction of travel is south, colour is black,
body type is saloon.Unhandled words:
Suspicious
Moira:
21:37
Confirm
there is a vehicle named 'v19' that has 'ABC123' as registration and has the vehicle body type 'saloon' as body type and has the colour 'black' as colour and is a moving thing
there is a moving thing named 'v19' that has the direction 'south' as direction of travel
Courtesy of Alun Preece and Dave Braines
open questions
∙ suitable fuzzy categories for representing uncertainty in a machine-to-humandialogue?
∙ ⟨で.ので, で.でと, で.どば⟩ ▶ possibly true (cf. Admiralty code)?∙ possibly true ▶ ⟨で.ので, で.でと, で.どば⟩ ?
∙ under which circumstances “quantities” can be translated into either assumptions orfacts ?
∙ water contaminated で x で y で z possible that water contaminated∙ water contaminated と で で で で で water contaminated
∙ plausibility metrics in interacting with human users?
white lies require coherence, but maybe up to a certain level?
∙ how to support querying provenance data?
∙ provenance is important enough to justify ad-hoc procedures∙ comparison with querying procedures for general S-DL-Lite KBs
open questions
∙ suitable fuzzy categories for representing uncertainty in a machine-to-humandialogue?
∙ で ので で でと で どば possibly true (cf. Admiralty code)?∙ possibly true で ので で でと で どば ?
∙ under which circumstances “quantities” can be translated into either assumptions orfacts ?
∙ water contaminated ⟨で.x, で.y, で.z⟩ ▶ possible that water contaminated∙ water contaminated ⟨と.で, で.で, で.で⟩ ▶ water contaminated
∙ plausibility metrics in interacting with human users?
white lies require coherence, but maybe up to a certain level?
∙ how to support querying provenance data?
∙ provenance is important enough to justify ad-hoc procedures∙ comparison with querying procedures for general S-DL-Lite KBs
open questions
∙ suitable fuzzy categories for representing uncertainty in a machine-to-humandialogue?
∙ で ので で でと で どば possibly true (cf. Admiralty code)?∙ possibly true で ので で でと で どば ?
∙ under which circumstances “quantities” can be translated into either assumptions orfacts ?
∙ water contaminated で x で y で z possible that water contaminated∙ water contaminated と で で で で で water contaminated
∙ plausibility metrics in interacting with human users?white lies require coherence, but maybe up to a certain level?
∙ how to support querying provenance data?
∙ provenance is important enough to justify ad-hoc procedures∙ comparison with querying procedures for general S-DL-Lite KBs
open questions
∙ suitable fuzzy categories for representing uncertainty in a machine-to-humandialogue?
∙ で ので で でと で どば possibly true (cf. Admiralty code)?∙ possibly true で ので で でと で どば ?
∙ under which circumstances “quantities” can be translated into either assumptions orfacts ?
∙ water contaminated で x で y で z possible that water contaminated∙ water contaminated と で で で で で water contaminated
∙ plausibility metrics in interacting with human users?
white lies require coherence, but maybe up to a certain level?
∙ how to support querying provenance data?∙ provenance is important enough to justify ad-hoc procedures∙ comparison with querying procedures for general S-DL-Lite KBs
ControlledQuery Evaluation
UncertainKB
NLInterfaces
S-DL-LiteKB ITA-CE
Strategies
representing uncertainty in machine-to-human dialogue?
plausibility metrics?
evaluationcomparing w.r.t.
ad-hoc procedures
acknowledgement
This research was sponsored by the U.S. Army Research Laboratory and the U.K.Ministry of Defence and was accomplished under Agreement Number4ばととNF-でね-な-でででと. The views and conclusions contained in this document arethose of the author(s) and should not be interpreted as representing the officialpolicies, either expressed or implied, of the U.S. Army Research Laboratory, theU.S. Government, the U.K. Ministry of Defence or the U.K. Government. The U.S.and U.K. Governments are authorized to reproduce and distribute reprints forGovernment purposes notwithstanding any copyright notation hereon.
..references and credits
Biskup, J., Kern-Isberner, G., Krümpelmann, P., and Tadros, C. (どでとに).Reasoning on Secrecy Constraints under Uncertainty to Classify Possible Actions.In FoIKS どでとに, pages ばの–ととね.
Braines, D., Preece, A., de Mel, G., and Pham, T. (どでとに).Enabling CoIST users: DどD at the network edge.In FUSION, pages と–は.
Şensoy, M., Fokoue, A., Pan, J. Z., Norman, T. J., Tang, Y., Oren, N., and Sycara, K. (どでとな).Reasoning about uncertain information and conflict resolution through trust revision.In AAMAS, pages はなの–はにに.
Idika, N., 3aria, M., and Phan, H. (どでとな).The Probabilistic Provenance Graph.In SP4, pages なに–にと.
Jøsang, A. (どででと).A Logic for Uncertain Probabilities.International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems, ば(な):どのば–なとと.
Preece, A., Braines, D., Pizzocaro, D., and Parizas, C. (どでとに).Human-machine conversations to support multi-agency missions.MCどR, とは(と):のぬ–はに.
credits
Template
adapted from mtheme https://github.com/matze/mtheme
Images
Lake: http://www.morguefile.com/archive/display/ぶどぢづぱで
Miles: http://www.morguefile.com/archive/display/ぱばびぴづだ
Ella: http://www.morguefile.com/archive/display/ででづびび
Sensor: http://www.morguefile.com/archive/display/ぱでづづぴぢ
NGO Laboratory: http://www.morguefile.com/archive/display/ぴづどぴばだ
UK Laboratory: http://www.morguefile.com/archive/display/ぱでぱぶでば